基于集成学习的信号控制交叉口排队长度估计

Queue Length Estimation at Signalized Intersection Based on Ensemble Learning

摘要:

基于电子警察（LPR）数据和网联车辆轨迹数据，提出了一种基于集成学习的信号控制交叉口排队长度估计方法。通过分析不同数据条件下估计方法的适用条件和精度水平，运用随机森林方法设计集成学习器，并构建电子警察和网联车辆轨迹感知信息及不同方法估计结果和真实排队长度之间的非线性映射关系。仿真结果表明：本方法的平均绝对误差为1.3 m?周期^-1?车道^-1，平均绝对百分比误差为1.4%。

Abstract:

Based on license plate recognition （LPR） data and connected vehicle trajectory data， an ensemble learning method was deployed to estimate the intersection queue length. By analyzing the applicability and accuracy of different queue length estimation methods， the random forest method was applied to design a base ensemble learner and formulize the nonlinear mapping relationship among the LPR data， connected vehicle trajectory data， estimation results of the existing queue length methods and real queue length values. Simulation results show that the proposed method overperforms the existing queue length methods， with a mean absolute error of 1.3 m?cycle^-1?lane^-1 and a mean absolute percent error of 1.4%.